Appendix A: The Claude Code Cheat Sheet

Print this. Tape it to the side of your monitor. Don’t try to memorize it; just know where to look.

Open / close / reset

The ten essential slash commands

Keystrokes worth knowing

Models, at a glance

The 8 built-in tools

The eight tools you’ll see most (Chapter 5 has the long version):

Read Write Edit Bash Glob Grep WebSearch WebFetch

There are a few more — notably Task, which spawns subagents (Chapter 13).

File layout

~/.claude/                           # personal, lives across all projects
├── CLAUDE.md                        # your global preferences
├── settings.json                    # global permissions + hooks
├── commands/                        # your personal slash commands
└── agents/                          # your personal subagents

<your-project>/
├── CLAUDE.md                        # project preferences
├── .claude/
│   ├── settings.json                # project allow/deny + hooks
│   ├── commands/                    # project slash commands
│   ├── agents/                      # project subagents
│   ├── skills/<name>/SKILL.md       # project skills
│   └── rules/                       # conditional rules

Permissions one-liner (.claude/settings.json)

{
  "permissions": {
    "allow": [
      "Read(*)",
      "Bash(npm test)",
      "Bash(pytest *)",
      "Bash(pre-commit run *)",
      "Bash(git status)",
      "Bash(git diff *)",
      "Bash(gh pr *)"
    ],
    "deny": [
      "Read(.env)",
      "Read(secrets/**)",
      "Read(data/patients/**)",
      "Read(*.key)",
      "Read(*.pem)"
    ]
  }
}

Auto-format hook (.claude/settings.json)

{
  "hooks": {
    "PostToolUse": [{
      "matcher": "Edit|Write",
      "hooks": [{
        "type": "command",
        "command": "jq -r '.tool_input.file_path' | xargs -r npx prettier --write 2>/dev/null || true"
      }]
    }]
  }
}

The hook receives a JSON description of the tool call on stdin; jq -r '.tool_input.file_path' pulls out the path of the file Claude just edited. (jq is a small command-line tool for reading JSON.)

Headless and scripting

# One-shot, no prompts
claude -p "fix the failing tests" --allowedTools Read,Edit,Bash

# Pipe a log file in
cat error.log | claude -p "explain this error and suggest a fix"

# Run three parallel worktrees on different branches
claude --worktree feature-auth
claude --worktree bugfix-login
claude --worktree refactor-api

Pricing (as of writing)

The starter CLAUDE.md

# CLAUDE.md
## Stack
<your stack in one sentence>

## Run / test
- Tests: <command>
- Eval: <command>
- Lint: <command>

## Behavior rules
1. Think before coding: verify, don't guess.
2. Simplicity first: minimum code that solves the problem.
3. Surgical changes: don't touch adjacent code.
4. Tests encode intent.
5. Fail loud: surface every skipped row, null, partial success.

## PHI rules (KHCC)
- MRNs: Optimus-encoded only in outputs.
- Names: Fernet-encrypted, never in logs.
- Eval cohort: frozen deceased-patient set only.

Daily flow

1. claude to open (or cc if you’ve configured the alias).
2. /statusline — one-time per laptop.
3. Shift+Tab into Plan Mode for anything bigger than a one-liner.
4. Read the plan; ask for changes before approving.
5. Watch the diff before approving each change.
6. /clear between unrelated tasks.
7. /cost before lunch; /cost again before you leave.

Glossary

A one-line definition for the vocabulary used in this book.

• Terminal — the window you type commands into. The room Claude Code lives in.
• LLM (large language model) — the AI engine that reads and writes text. Claude is one.
• Agent — an LLM given tools and a loop: it reads, acts, checks the result, and repeats until the job is done.
• Token — the unit models read and bill in; roughly three-quarters of an English word.
• Context window — the model’s working memory for one conversation. When it fills, older detail is summarized or lost.
• Prompt (in the terminal) — the symbol that says “ready for your next command” ($, >, or %). Not to be confused with a prompt to the model, which is the sentence you type to Claude.
• Command — the verb you type at the terminal (ls, git, claude).
• Argument — the noun after the verb (cd Documents — Documents is the argument).
• Flag — a modifier on a command, written with one or two dashes (ls -l, claude --resume).
• Directory — a folder. The terminal calls folders “directories” because it’s old.
• Repo (short for repository) — a folder tracked by git for version history.
• Branch — a parallel timeline of changes in a repo. You can have many. main is the default.
• Commit — a snapshot of the repo at a moment in time, with a message describing what changed.
• Push — send your commits from your laptop up to a remote server (e.g., GitHub).
• Pull — fetch other people’s commits from the remote into your local copy.
• Allowlist — the list of commands Claude can run without asking, written in .claude/settings.json.
• Deny list — the list of paths Claude is forbidden to read or write.
• Slash command — a Claude Code shortcut typed with a leading / (e.g., /clear).
• Hook — a small program Claude Code runs automatically at certain moments (before a tool, after a tool, at session end).
• Checkpoint — an automatic snapshot of your files taken as Claude works, separate from git. Restore one with /rewind.
• Plan Mode — a permission mode (reached with Shift+Tab) where Claude proposes a plan and edits nothing until you approve.
• Subagent — a separate Claude instance the main session can spawn for a self-contained job; has its own context, tools, and model.
• MCP (Model Context Protocol) — a standard way for Claude to talk to external tools and databases via small servers.
• Worktree — a separate working copy of the same repo, on a different branch, so you can run two sessions in parallel.
• Headless mode — running Claude Code non-interactively with claude -p, for scripts and scheduled jobs.
• Ralph loop — a pattern that keeps relaunching Claude on a task until a completion promise is met; always guarded by a max-iterations limit.
• CLAUDE.md — the project’s instruction file; Claude reads it every session.
• Skill — a packaged capability under .claude/skills/<name>/SKILL.md that Claude can invoke.
• PHI — Protected Health Information; anything that identifies a patient.
• Fernet — symmetric encryption used at the AI Office to protect patient names.
• Optimus encoding — reversible scrambling of MRNs into non-human-readable identifiers.
• Eval cohort — at the AI Office, the frozen list of deceased patients with hand-curated ground truth, used to grade every prompt change.